COS 429 - Computer Vision

Fall 2019

Assignment 2: Face Detection and Model Fitting

Changelog and clarifications

• 10/15/19: Part V f): Both understandings (i.e., 10 inliers or 100 inliers) will get full credit.
• 10/10/19: Clarified the plots needed in Part I c). Only plots of logistic regression are needed.
• 10/09/19: Clarified the number of training samples in Part II c), d), and e). The classifier should be trained on 6,000 faces and 6,000 nonfaces, and tested on 500 faces and 500 nonfaces.
• 10/07/19: Submission instructions posted.

1. Logistic regression (10 pts)
2. Training a face classifier (30 pts)
3. Single-scale face detection (20 pts)
4. Multi-scale face detection (10 pts)
5. RANSAC and Hough transform (30 pts)

Submitting:

This assignment is due Thursday, October 17, 2019 at 11:59 PM. Please see the general notes on submitting your assignments, as well as the late policy and the collaboration policy.

As stated, our submissions this year will be done through Gradescope. The code to register for our class and how to submit can be found in the "general notes on submitting..." link above. This assignment has 2 submissions:

1. Assignment 2 Written: Submit one single PDF containing all written portions of the assignments. Everything other than code should be in this PDF file. If you attempt extra credit, specify within the writeup where the code for the extra credit can be found.
2. Assignment 2 Code: Submit all your code. At least the following 7 files should be submitted: find_faces_single_scale.py, find_faces.py, get_testing_data.py, get_training_data.py, logistic_fit.py, logistic_prob.py, hog36.py. Additional code from extra credit can be submitted as well, as long as it is specified in the written portion which extra credit questions correspond to the extra code. We will not use autograder this time, but will read your code and grade it manually.

Note that:

1. Assignment 2 Written is worth 100 points (actually 110 with extra credits) and Assignment 2 Code is worth 0 points on gradescope. This does not indicate we do not grade the coding part; we just put all the gradings in the written part on gradescope.
2. For extra credit, you have to clearly state where your code is, if your answer contains code.